Gregg Richard E, Fiol-Sala Miquel, Nikus Kjell C, Startt-Selvester Ronald, Zhou Sophia H, Carrillo Andrés, Barbara Victoria, Chien Cheng-hao Simon, Lindauer James M
Advanced Algorithm Research Center, Philips Healthcare, Thousand Oaks, CA, USA.
J Electrocardiol. 2012 Jul-Aug;45(4):343-9. doi: 10.1016/j.jelectrocard.2012.03.008.
Classifying the location of an occlusion in the culprit artery during ST-elevation myocardial infarction (STEMI) is important for risk stratification to optimize treatment. We developed a new logistic regression (LR) algorithm for 3-group classification of occlusion location as proximal right coronary artery (RCA), middle-to-distal RCA or left circumflex (LCx) coronary artery with inferior myocardial infarction. We compared the performance of the new LR algorithm with the recently introduced decision tree classifier of Fiol et al (Ann Noninvasive Electrocardiol. 2004;4:383-388) in the classification of the same 3 categories.
The new algorithm was developed on a set of electrocardiograms from an emergency department setting (n = 64) and tested on a different set from a prehospital setting (n = 68). All patients met the current STEMI criteria with angiographic confirmation of culprit artery and occlusion location. Using LR, 4 ST-segment deviation features were chosen by forward stepwise selection. Final LR coefficients were obtained by averaging more than 200 bootstrap iterations on the training set. In addition, a separate 4-feature classifier was designed adding ST features of V4R and V8, only available in the training set.
The LR algorithm classified proximal RCA occlusion vs combined LCx occlusion and middle-to-distal RCA occlusion, with a sensitivity of 76% and specificity of 81% as compared with 71% and 62% for the Fiol classifier. The difference in specificity was statistically significant. The LR classifier trained with additional ST features of V4R and V8, but still limited to 4, improved the overall agreement in the training set from 65% to 70%.
Discrimination of proximal RCA lesion location from LCx or middle-to-distal RCA using the new LR classifier shows improvement over decision tree–type classification criteria. Automated identification of proximal RCA occlusion could speed up the risk stratification of patients with STEMI. The addition of leads V4R and V8 should further improve the automated classification of the occlusion site in RCA and LCx.
对ST段抬高型心肌梗死（STEMI）罪犯血管闭塞部位进行分类对于风险分层以优化治疗非常重要。我们开发了一种新的逻辑回归（LR）算法，用于对伴有下壁心肌梗死的闭塞部位进行三组分类，即右冠状动脉近端（RCA）、RCA中远端或左旋支（LCx）冠状动脉。我们将新LR算法的性能与Fiol等人最近引入的决策树分类器（《非侵入性心电图学杂志》。2004年；4：383 - 388）在相同三类别的分类性能进行了比较。
新算法基于一组急诊科心电图（n = 64）开发，并在另一组院前心电图（n = 68）上进行测试。所有患者均符合当前STEMI标准，并有罪犯血管和闭塞部位的血管造影确认。使用LR，通过向前逐步选择选择了4个ST段偏移特征。最终的LR系数通过对训练集进行200多次自助重抽样迭代平均得到。此外，设计了一个单独的4特征分类器，加入了仅在训练集中可用的V4R和V8的ST特征。
LR算法对RCA近端闭塞与LCx合并闭塞及RCA中远端闭塞进行分类，灵敏度为76%，特异性为81%，而Fiol分类器的灵敏度和特异性分别为71%和62%。特异性差异具有统计学意义。使用V4R和V8的额外ST特征训练的LR分类器，尽管仍限制为4个特征，但将训练集中的总体一致性从65%提高到了70%。
使用新的LR分类器区分RCA近端病变部位与LCx或RCA中远端病变部位，相较于决策树类型的分类标准有改进。自动识别RCA近端闭塞可加快STEMI患者的风险分层。加入V4R和V8导联应进一步改善RCA和LCx闭塞部位的自动分类。
